determine the impact of seedPerwindowNmax on samples

[slsevilla]

While processing RBL samples, errors were encountered relating to the seedPerwindowNmax value (set at 10000). To determine the impact of this value on multi-mapping stats, two samples were run at iterations.

Samples

• Ro7hr2_Clip
• YKO7hr_Clip

seedPerwindowNmax values

• 10
• 50 (default value)
• 1000
• 5000
• 10000 ( settings used)

Example code:

#!/bin/bash
module load STAR; \
sample_id="Ro7hr2_Clip"; \
seedPerWindowNmax="10";\
sample_id_prefix="${sample_id_prefix}_"; \
tmp_dir="/lscratch/${SLURM_JOB_ID}"; \
export tmp_dir; \
STAR --runMode alignReads --genomeDir /data/CCBR_Pipeliner/iCLIP/index/active/2022_0505/mm10/index \
--sjdbGTFfile /data/CCBR_Pipeliner/iCLIP/index/active/2022_0505/mm10/ref/gencode.vM23.annotation.gtf --readFilesCommand zcat \
--readFilesIn /data/RBL_NCI/Wolin/mESC_clip_4_v2.0/01_preprocess/01_fastq/$sample_id.fastq.gz \
--outFileNamePrefix $tmp_dir/${sample_id_prefix} \
--outReadsUnmapped Fastx --outSAMtype BAM SortedByCoordinate --alignEndsType Local --alignIntronMax 50000 --alignSJDBoverhangMin 3 \
--alignSJoverhangMin 5 --alignTranscriptsPerReadNmax=10000 --alignWindowsPerReadNmax=10000 --outFilterMatchNmin 15 --outFilterMatchNminOverLread 0.9 \
--outFilterMismatchNmax 999 --outFilterMismatchNoverReadLmax 0.04 --outFilterMultimapNmax 10000 --outFilterMultimapScoreRange 0 --outFilterScoreMin 0 \
--outFilterType Normal --outSAMattributes All --outSAMunmapped None --outSJfilterCountTotalMin 3 1 1 1 --outSJfilterOverhangMin 30 12 12 12 \
--outSJfilterReads All --seedMultimapNmax 10000 --seedNoneLociPerWindow 20 --seedPerReadNmax 10000 --seedPerWindowNmax $seedPerWindowNmax --sjdbScore 2 --winAnchorMultimapNmax 10000

 # move STAR files and final log file to output
 mv $tmp_dir/${sample_id_prefix}Aligned.sortedByCoord.out.bam /data/sevillas2/star/$seedPerWindowNmax/${sample_id}.bam
 mv $tmp_dir/${sample_id_prefix}Log.final.out /data/sevillas2/star/$seedPerWindowNmax/${sample_id}.out

 # move mates to unmapped file
 touch /data/sevillas2/star/$seedPerWindowNmax/${sample_id}.unmapped.out
 for f in $tmp_dir/${sample_id_prefix}Unmapped.out.mate*; do cat $f >> /data/sevillas2/star/$seedPerWindowNmax/${sample_id}.unmapped.out; done

[slsevilla]

Overview

• Testing Control sample using varied seed and anchor parameters:
• Seed values tested: 10, 50 (default), 100, 500, 1000, 10000 (used during STAR optimization)
• Anchor values tested: 50 (default),10000 (used during STAR optimization)
• Pairs tested (anchor/seed): 50/1000, 50/500, 50/100, 50/50, 10000/5000,10000/1000,10000/500,10000/100.10000/50
• Pairs NOT tested (anchor/seed): 50/5000, 50/10

Analysis

Unique reads

• The percentage of unique reads in the sample does not vary between anchor values when seed is set between 50-1000.
• The percentage of unique reads falls significantly (from ~76% to ~17%) when the seed is dropped from 50 to 10.

Spliced TOTAL number of reads

• The number of spliced reads falls (from ~2900 to ~1600) when the anchor is dropped from 10000 to 50.
• The number of spliced reads falls (from ~2900 to ~1900) when the seed is dropped from 50 to 10.

Spliced ANNOTATED number of reads

• There is no change in the number of annotated reads for any variation

Multi-mapped reads

• The percent of multimapped reads is lower (from ~4.3% to ~4%) in samples when the anchor is dropped from 10000 to 50.
• The percent of multimapped reads is lower (from ~4.24% to ~3.6%) in samples when the seed is dropped from 50 to 10.

Unmapped reads - unmapped due to 'too many reads in windows'

• There is no change in the number of reads for any variation.

• Unmapped reads - unmapped due to 'too short'

• The percentage of reads failing in the sample does not vary between anchor values when seed is set between 50-1000.
• The percentage of reads failing falls (from ~19% to ~16%) when the seed is dropped from 50 to 10.

Unmapped reads - unmapped due to 'other'

• The percentage of reads failing in the sample does not vary between anchor values when seed is set between 50-1000.
• The percentage of reads failing increases (from ~1% to ~64%) when the seed is dropped from 50 to 10.

Recommendation

The recommendation for the parameters are as follows -

• seedPerWindowNmax: 500
• winAnchorMultimapNmax: 50

These values will allow us to align samples without overstretching our resources quickly, while still maintaining a high number of unique reads, and annotated splice junctions.